Update README.md

README.md

@@ -1,15 +1,63 @@
+---
+title: 👗 Atelier d'AI
+emoji: ✨
+colorFrom: yellow
+colorTo: indigo
+sdk: docker
+pinned: true
+tags:
+- fashion
+- computer-vision
+- clip
+- quiet-luxury
+- vector-search
+---
+
 # 👗 Atelier d'AI: A Neural Fashion Search Engine
+### *An Ode to High Fashion, Paris Runway, and the 'Quiet Luxury' Aesthetic*
+
+**Atelier d'AI** is a sophisticated recommendation system designed for the "fashion girlies" who appreciate the timeless elegance of Dior, Chanel, and Louis Vuitton. Unlike traditional search engines that rely on keywords, this project utilizes **Deep Metric Learning** to understand the "DNA" of a garment—its silhouette, texture, and vibe.
+
+---
+
+## 🚀 The Technical Vision
+To move from "Emily in Paris" kitsch to high-fashion sophistication, the system moves beyond simple color matching. It uses a **two-stage AI pipeline**:
+
+1.  **Semantic Encoding:** Leveraging **Marqo-FashionSigLIP**, a Vision Transformer (ViT) fine-tuned on luxury fashion datasets. It maps images and text into a shared 768-dimensional vector space.
+2.  **Vector Retrieval:** Utilizing **FAISS (Facebook AI Similarity Search)** for lightning-fast nearest neighbor retrieval.
+
+[Image of a deep learning-based fashion recommendation system architecture including feature extraction and similarity matching]
+
+---
+
+## 🛠️ Tech Stack & Architecture
+* **Neural Backbone:** `OpenCLIP` + `Marqo-FashionSigLIP` (Sigmoid Loss-Image Language Pre-training)
+* **Vector Database:** `FAISS` (IndexFlatIP for Cosine Similarity)
+* **Infrastructure:** `Docker` (ensuring environment parity for production)
+* **Frontend:** `Streamlit` with a custom "Luxury Dark" CSS theme
+
+---
+
+## 💎 Features
+* **Visual Inspiration Search:** Upload a Pinterest moodboard or a runway snap to find pieces with a similar "Quiet Luxury" silhouette.
+* **Aesthetic Prompting:** Search using abstract high-fashion concepts like *"90s minimalist chic"* or *"Parisian street style"* using CLIP's zero-shot capabilities.
+* **Metric-Driven Results:** Every recommendation includes a confidence score based on vector distance.
+
+---
+
+## 🧠 Technical Deep Dive: Why Vector Search?
+Traditional search looks for "tags." Atelier d'AI looks for **relationships**. 
+
+By performing **L2 Normalization** on our 768-d embeddings, we ensure that similarity is measured by the *direction* of the style vector. This means the model understands that a "tweed texture" is a style signature, regardless of the photo's brightness or the garment's color.
+
+$$\text{Cosine Similarity} = \frac{A \cdot B}{\|A\|\|B\|}$$
+
+This mathematical approach allows for **Zero-Shot Retrieval**, where the model can find "Quiet Luxury" items even if it has never been explicitly told what that label means.
 
-### The Vision
-An ode to the "fashion girlies" and the timeless elegance of Chanel, Dior, and LV. This engine uses **Deep Metric Learning** to understand style silhouettes beyond simple keywords.
+---
 
-### Tech Stack
-- **Model:** Marqo-FashionSigLIP (Vision Transformer)
-- **Engine:** OpenCLIP + PyTorch
-- **Vector DB:** FAISS (Facebook AI Similarity Search)
-- **UI:** Streamlit in a Docker Container
+## 🤝 Contributing & License
+Distributed under the **MIT License**. Created as a portfolio piece to demonstrate the intersection of **Computer Vision** and **Luxury Aesthetics**.
 
-### How to use
-1. Upload a runway photo or a Pinterest moodboard.
-2. The AI extracts a 768-dimensional "Style Vector".
-3. FAISS performs a Nearest Neighbor search to find the most elegant match.
+**Developed by Priyanshi Shah**
+*Student at PDEU | Aspiring AI/ML Leader*